Playing10 min read

Fried Liver Attack: What Each Rating Actually Plays

The Fried Liver Attack only happens if Black takes on d5 at move five. Chessdrive's Maia3 and Stockfish data shows how that choice flips between 1400 and 1600, how often White finds 6.Nxf7 and the quiet 7.Qf3+, and why the Lolli 6.d4 is almost never played.

The starting position

Five moves into a Two Knights game, one capture decides whether the Fried Liver Attack happens at all. After 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5, Black either plays 5...Nxd5 and allows 6.Nxf7, or plays something else and never sees it. We ran the maia3-5m checkpoint, see how Maia chess actually works, on four positions along this line at seven rating bands from 1000 to 2200, and scored every legal move with Stockfish 18 at depth 20, see what the Stockfish chess engine is. This is the same method as our chess openings by rating dataset. Maia3 predicts what a player at a given rating is likely to choose. It is a model, not a count of real games. We tried to add real game counts from the Lichess explorer, and it was unavailable, so none appear here.

One note on signs. Stockfish reports a score from the side to move. Every evaluation below is written from White's point of view: +0.27 means White is ahead by about a quarter of a pawn, and a negative number means Black is better.

The move order and why it works

The full line is 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5 Nxd5 6.Nxf7. It is a Two Knights Defense variation, the sort of line that sits inside the Italian Game family, and Wikipedia gives it ECO code C57. White gives up a knight for a pawn, and after 6...Kxf7 the black king has left home. White's queen then checks on f3 and the main line sends the king toward the centre with 7...Ke6, where it defends the d5 knight. The main line runs 7.Qf3+ Ke6 8.Nc3 Nb4 9.0-0, where 9.0-0 is the move both Leela and Stockfish chose in a pair of engine games recorded in a chess.com blog post (engine versions not stated), and both games were drawn.

Whether the sacrifice is sound is a real argument. As of November 2026, Wikipedia says White has a strong attack that has not yet been proven decisive. One chess.com forum poster quotes an engine at roughly +1.1 for 6.Nxf7, with no engine, depth or date given, so we do not rely on it. Our own Stockfish 18 depth 20 run puts the sacrifice at +0.92 for White, a clear but not crushing edge. The attack has always been more dangerous than it looks on the board, because the defence is harder to find than the attack. Wikipedia notes that Black does better in slow games, where there is time to find the refutations, and that is why the line is most dangerous at short time controls.

The name has an Italian root. Fegatello is a dish of pork liver, sliced, wrapped in fat netting and cooked over a fire, and Fried Liver is the English rendering. Wikipedia gives the earliest known example as a Giulio Cesare Polerio game. Its Fried Liver page says before 1606 and its Two Knights page says around 1580, so the date is uncertain, and all we can safely say is the late 1500s or early 1600s. Tarrasch had little time for 4.Ng5, calling it a duffer's move (ein richtiger Stümperzug in German, translated as bungling move or beginner's move in different places). The Fried Liver is the reason people play it anyway.

Move 4: 4...d5 is the likeliest move at every band

After 4.Ng5, Black needs to deal with the threat on f7. Stockfish ranks 4...d5 first at +0.27 for White, with 4...Bc5 second at +1.17 and 4...Nxe4 at +2.14. Wikipedia describes 4...Bc5 as the Traxler Counterattack and says it leads to wild complications. The engine thinks White is clearly better after it, though forum posters and some repertoire sites treat it as a real weapon. Treat it as sharp and risky, not as the standard reply.

Rating1000120014001600180020002200
Probability of 4...d5 (best, White +0.27)68.3%76.5%77.0%76.5%76.7%76.8%78.2%
Probability of 4...Bc5, the Traxler (White +1.17)7.7%9.4%12.3%13.9%14.4%15.4%16.6%
Probability of 4...Nxe4 (White +2.14)5.1%4.9%5.6%6.4%6.9%6.2%4.0%

The engine's move is also the likeliest human move at every band, from 68.3 percent at 1000 to 78.2 at 2200. The Traxler is the move that grows with rating, from 7.7 percent at 1000 to 16.6 at 2200, rising at every band. 4...Qe7 is the odd one at the bottom: it takes 8.7 percent at 1000, second behind 4...d5 there, and falls to 3.3 at 1200 and 1.2 at 1400. It is scored at +1.99 for White. So Black almost always reaches 4...d5, and the fork comes one move later.

Move 5: where the crowd flips

After 5.exd5 Black has the choice the whole opening hangs on. Stockfish puts 5...Na5 first at +0.18 for White, 5...b5 second at +0.71, 5...Nd4 third at +0.77 and 5...Nxd5 fifth at +0.97. Wikipedia calls 5...Na5 generally considered best and names it the Chigorin Gambit; forum sources call it the Polerio Defense. Wikipedia also says 5...Nxd5 is considered too risky at the board but has not been definitively refuted, which fits a score of under a pawn rather than a lost game.

Rating1000120014001600180020002200
Probability of 5...Nxd5, allows the Fried Liver (White +0.97)81.2%70.5%52.4%35.2%21.7%11.9%6.3%
Probability of 5...Na5 (best, White +0.18)10.0%20.3%35.7%49.2%59.1%65.7%69.7%
Probability of 5...Nd4 (White +0.77)5.2%6.0%7.7%9.3%10.1%9.6%8.2%
Probability of 5...b5 (White +0.71)0.5%1.0%2.6%5.1%8.4%12.3%15.6%

This is the cleanest pattern in the data. At 1000 Maia3 predicts 5...Nxd5 at 81.2 percent, and only 10.0 for 5...Na5. The two cross between 1400 and 1600: at 1400 the capture still leads, 52.4 to 35.7, and at 1600 the knight move leads, 49.2 to 35.2. By 2200 it is 69.7 percent for 5...Na5 and 6.3 for the capture. Every step is in one direction, so the sacrifice is a feature of lower rated games and mostly vanishes above 2000.

The second story is how the alternatives grow. 5...b5, the Ulvestad, climbs from 0.5 percent at 1000 to 15.6 at 2200 and overtakes 5...Nxd5 at 2000 (12.3 against 11.9). 5...Nd4, the Fritz, sits at 5.2 to 10.1 percent and peaks at 1800. Wikipedia notes that these two usually transpose into each other, so a 2000 player who plays either is often choosing a different route to the same family of positions. All three of the quiet moves score between +0.18 and +0.77, which is why a player can pick the one they like. The capture 5...Nxd5 is the one that costs the most, and also the one that leads to the attack.

What White should do at move 6

If you are White and Black has taken on d5, the engine's choice is 6.Nxf7 at +0.92. That is the likeliest move at every band, but the share changes a lot.

Rating1000120014001600180020002200
Probability of 6.Nxf7 (best, +0.92)31.8%42.8%54.7%65.7%73.7%78.1%78.8%
Probability of 6.Qf3 (-0.77)27.4%23.0%19.2%15.1%11.5%9.2%8.4%
Probability of 6.d3 (-0.14)15.9%17.2%14.0%9.8%6.7%4.8%4.2%
Probability of 6.d4, the Lolli (+0.79)5.6%5.3%4.7%4.3%4.5%5.2%6.1%
Probability of 6.Bxd5 (-1.86)8.2%5.0%3.4%2.4%1.6%1.0%out

In this table, "out" means the move was outside Maia's top five at that band, not zero probability. At 1000 only about a third of predicted White moves are the sacrifice. Another 27.4 percent go to 6.Qf3, which the engine scores at -0.77, meaning Black is better. This is the common wrong try: it keeps the knight and threatens f7, and it hands Black the advantage. By 2200 the sacrifice takes 78.8 percent and 6.Qf3 is down to 8.4. The quiet 6.d3, scored at -0.14, takes 17.2 percent at 1200 and fades after that.

The Lolli stands out. The Wikipedia page says Fischer felt 6.d4 was incredibly strong, and our engine agrees it is nearly as good as the sacrifice, at +0.79 against +0.92. But Maia3 gives it only 5.6 percent at 1000, 4.3 at 1600 and 6.1 at 2200. It never exceeds 6.1 percent at any band, even though it is within 0.13 of the best move. If you play White and want to take Black out of preparation, the engine thinks 6.d4 is a respectable second weapon. We have no real game results to say how it scores in practice, so treat that as an engine claim, not a practical one.

After 6...Kxf7, one move only

Once the knight is gone and Black has recaptured, White has one good move. The engine scores 7.Qf3+ at +0.76. The next best move, 7.Nc3, is -2.82, and the others sit at about -2.86 or worse. That is a swing from +0.76 to -2.82 on a single slip, which is why this position punishes anyone who learned the sacrifice without the follow up.

Rating1000120014001600180020002200
Probability of 7.Qf3+ (only move, +0.76)95.1%96.8%97.4%97.8%98.0%98.0%97.7%
Probability of 7.Qh5+ (-3.48)3.1%2.1%1.8%1.4%1.0%0.7%0.5%

Nearly everyone finds it. The model gives 7.Qf3+ 95.1 percent at 1000 and 98.0 at 1800 and 2000, and the only visible slip at the low end is 7.Qh5+ at 3.1 percent, which the engine scores at -3.48. At 2200, 7.Nc3 takes 1.7 percent. The practical reading is that the follow up is easy and the decision that matters is move 6, where lower rated White players often pick 6.Qf3 over 6.Nxf7. A White player who memorises 6.Nxf7 and 7.Qf3+ has memorised the part the engine requires.

What this means for each colour

As Black, the data argues for 5...Na5. At 1000 Maia3 predicts the capture on d5 81.2 percent of the time, and at 1400 it is still more than half, so this is the move where lower rated players are most likely to be walked into a prepared attack. Stockfish says the capture is only +0.97 for White, not a loss, but the engine does not play the position for you. The attacker has the initiative and the defence needs accuracy. For the wider family this line belongs to, our Italian Game rating by rating shows what each band plays at move three.

As White, 4.Ng5 is a specialist's move, and the data shows why it works best at lower ratings. The trap needs Black to take on d5, and from 1600 up that capture becomes much rarer: 35.2 percent at 1600, 21.7 at 1800, 6.3 at 2200. Above that level you are choosing between an equalish line after 5...Na5 and something quieter. Learn the plan rather than the moves, as argued in why learning plans beats memorizing openings. A gambit with a real structural idea, like the Evans Gambit, travels to higher ratings in a way that a single sacrifice does not. For how the Fried Liver compares with six other opening traps band by band, see our chess traps data.

Where other pages are better

Wikipedia is the better source for the history and for the full list of named defences, and it gives ECO codes and the correspondence chess note. The chess.com opening page is the better introduction for a beginner who wants the story, including the Morphy game and a 2600 level game. ChessKid is aimed at young players and walks through the attack in a friendlier way. Chess.com forum threads and blogs are where you will find long arguments on whether the sacrifice is sound, and several game by game defence walkthroughs. A few of those, though, have garbled move numbers and no engine numbers, so they can mislead. Two further pages, thechessworld and a GM article on chess.com, are likely fuller guides than this one, but we could not read them in full.

What none of those pages shows is what each rating band actually chooses, or an engine score on every legal move. That is the gap here. What this page does not cover: real game results, the Lichess explorer counts, the deep lines after 8...Nb4 9.0-0, and how the Traxler plays out beyond move four.

What to take from it

The Fried Liver is a lower rating phenomenon. 4...d5 is the likeliest move at every band, the capture on d5 is the likeliest move up to 1400 and the retreat 5...Na5 is the likeliest from 1600, and from 1000 to 2200 nearly everyone as White finds 7.Qf3+ once the sacrifice is on the board. The real mistake is at move six, where 6.Qf3 is the common wrong try and 6.d4 is the underused good one. The sacrifice is a good practical weapon and a modest engine edge.

Maia3 is a model of human choice built from the maia3-5m checkpoint, not a count of real games, and every percentage above should be read that way.

Frequently asked

What is the Fried Liver Attack?
It is a Two Knights Defense line in which White sacrifices a knight on f7 to drag the Black king out. The move order is 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5 Nxd5 6.Nxf7. Wikipedia files it under ECO C57 and also calls it the Fegatello Attack, after an Italian dish of pork liver.
Is the Fried Liver Attack good for White?
Yes, but only slightly by the engine. In our data Stockfish 18 at depth 20 scores 6.Nxf7 at +0.92 for White, and 5...Nxd5 at +0.97 for White before the sacrifice. Wikipedia says White has a strong attack that has not been proven decisive, and a chess.com blog post (engine versions not stated) reports that Leela and Stockfish both chose 9.0-0 in the critical line and both games were drawn. Over the board, the attack is far more dangerous in fast games than in slow ones.
How do you defend against the Fried Liver Attack?
Do not take on d5. After 4.Ng5 d5 5.exd5, the move 5...Na5 is generally considered Black's best fifth move. Stockfish puts it at +0.18 for White, against +0.97 after 5...Nxd5. Wikipedia lists 5...b5 (Ulvestad) and 5...Nd4 (Fritz) as other named tries that usually transpose into each other. If you have already played 5...Nxd5 and White sacrifices, the main line is 6...Kxf7 7.Qf3+ Ke6.
Why is it called the Fried Liver Attack?
The Italian name is Fegatello, which Wikipedia explains as sliced pork liver wrapped in fat netting and cooked over a fire. The English name is a translation of it. Another site gives a separate idiom about liver as the origin, but only the dish is confirmed in the sources we checked.
What is the Lolli Attack, and is it better than 6.Nxf7?
The Lolli Attack is 6.d4 instead of 6.Nxf7. Wikipedia says Bobby Fischer felt it was incredibly strong, to the point that 5...Nxd5 is rarely played. Stockfish scores it at +0.79, close behind 6.Nxf7 at +0.92. Maia3 gives 6.d4 between 4.3% and 6.1% at every rating we measured, so it is the engine close move that almost nobody plays.
Who invented the Fried Liver Attack?
The earliest known example is a game by Giulio Cesare Polerio. Wikipedia's two pages disagree on the date: one says before 1606, the other says around 1580. We would only say the late sixteenth or early seventeenth century.

Want to know if your own rating walks into 5...Nxd5?

Chessdrive runs the real maia3-5m checkpoint on-device as your opponent at any rating you set from 600 to 2600, so you can play 4.Ng5 and find out whether you defend with 5...Na5 or grab the pawn under a clock. Finished games are reviewed separately with Stockfish so you can see exactly where the position turned. Available now on iOS.

Keep reading

All guides

Updated